copies with memcpy instead of memmove. This matches what GCC does and if it
causes a problem with a particular libc we can always fix it with a target
hook.
llvm-svn: 65699
- When we are declaring a function in local scope, we can merge with
a visible declaration from an outer scope if that declaration
refers to an entity with linkage. This behavior now works in C++
and properly ignores entities without linkage.
- Diagnose the use of "static" on a function declaration in local
scope.
- Diagnose the declaration of a static function after a non-static
declaration of the same function.
- Propagate the storage specifier to a function declaration from a
prior declaration (PR3425)
- Don't name-mangle "main"
llvm-svn: 65360
- PR3463, PR3398, <rdar://problem/6553401> crash on relocatable
symbol addresses as constants in static locals.
- There are many more scenarious we could handle (like arithmetic on
such an int) but this is the main use case.
llvm-svn: 65074
situation where a tentative decl was emitted *after* the actual
initialization. This occurs in some rare situations with static decls.
- PR3613.
- I'm not particularly happy with this fix, but I don't see a simpler
or more elegant solution yet.
llvm-svn: 65018
- Renamed to getDeclAlignInBytes since most other query functions
work in bits.
- Fun to track down as isIntegerConstantExpr was getting it right,
but Evaluate() was getting it wrong. Maybe we should assert they
compute the same thing when they succeed?
llvm-svn: 64828
IRgen no longer relies on isConstantInitializer, instead we just try
to emit the constant. If that fails then in C we emit an error
unsupported (this occurs when Sema accepted something that it doesn't
know how to fold, and IRgen doesn't know how to emit) and in C++ we
emit a guarded initializer.
This ends up handling a few more cases, because IRgen was actually
able to emit some of the constants Sema accepts but can't Evaluate().
For example, PR3398.
llvm-svn: 64780
- Define pow[lf]?, sqrt[lf]? as builtins.
- Add -fmath-errno option which binds to LangOptions.MathErrno
- Add new builtin flag Builtin::Context::isConstWithoutErrno for
functions which can be marked as const if errno isn't respected for
math functions. Sema automatically marks these functions as const
when they are defined, if MathErrno=0.
- IRgen uses const attribute on sqrt and pow library functions to
decide if it can use the llvm intrinsic.
llvm-svn: 64689
etc.) when we perform name lookup on them. This ensures that we
produce the correct signature for these functions, which has two
practical impacts:
1) When we're supporting the "implicit function declaration" feature
of C99, these functions will be implicitly declared with the right
signature rather than as a function returning "int" with no
prototype. See PR3541 for the reason why this is important (hint:
GCC always predeclares these functions).
2) If users attempt to redeclare one of these library functions with
an incompatible signature, we produce a hard error.
This patch does a little bit of work to give reasonable error
messages. For example, when we hit case #1 we complain that we're
implicitly declaring this function with a specific signature, and then
we give a note that asks the user to include the appropriate header
(e.g., "please include <stdlib.h> or explicitly declare 'malloc'"). In
case #2, we show the type of the implicit builtin that was incorrectly
declared, so the user can see the problem. We could do better here:
for example, when displaying this latter error message we say
something like:
'strcpy' was implicitly declared here with type 'char *(char *, char
const *)'
but we should really print out a fake code line showing the
declaration, like this:
'strcpy' was implicitly declared here as:
char *strcpy(char *, char const *)
This would also be good for printing built-in candidates with C++
operator overloading.
The set of C library functions supported by this patch includes all
functions from the C99 specification's <stdlib.h> and <string.h> that
(a) are predefined by GCC and (b) have signatures that could cause
codegen issues if they are treated as functions with no prototype
returning and int. Future work could extend this set of functions to
other C library functions that we know about.
llvm-svn: 64504
- Fix emission of static functions with constructor attribute while I
was here.
<rdar://problem/6140899> [codegen] "static" and attribute-constructor interact poorly
llvm-svn: 64488
ABI to the CodeGen library. Since C++ code-generation is so
incomplete, we can't exercise much of this mangling code. However, a
few smoke tests show that it's doing the same thing as GCC. When C++
codegen matures, we'll extend the ABI tester to verify name-mangling
as well, and complete the implementation here.
At this point, the major client of name mangling is in the uses of the
new "overloadable" attribute in C, which allows overloading. Any
"overloadable" function in C (or in an extern "C" block in C++) will
be mangled the same way that the corresponding C++ function would be
mangled.
llvm-svn: 64413
represents an implicit value-initialization of a subobject of a
particular type. This replaces the (ab)use of CXXZeroValueInitExpr
within initializer lists for the "holes" that occur due to the use of
C99 designated initializers.
The new test case is currently XFAIL'd, because CodeGen's
ConstExprEmitter (in lib/CodeGen/CGExprConstant.cpp) needs to be
taught to value-initialize when it sees ImplicitValueInitExprs.
llvm-svn: 63317
- Although gross, this is needed currently to ensure that we produce
well formed debug information (to match pace with the assertions
being added to DebugInfo in LLVM).
llvm-svn: 62734
When emitting the static variables we need to make sure that the order is preserved.
Fix this by making StaticDecls a std::list which has O(1) random removal.
llvm-svn: 61621
simple test that actually does VLA codegen.
Note that despite the fact that the alloca isn't in the entry block, it
should dominate all uses; this is guaranteed by the restrictions on goto
into VLA scope in C99.
llvm-svn: 61291
value).
- Use extra argument to EmitStoreThroughLValue to provide place to
write update bit-field value if caller requires it.
- This fixes several FIXMEs.
llvm-svn: 59615
aliases.
- Attributes specific to a definition are only set when the
definition is seen.
- Alias generation is delayed until the end of the module; necessary
since the alias may reference forward.
- Fixes: PR2743, <rdr://6140807&6094512>
- Improves: <rdr://6095112> (added XFAIL)
Also, print module on verification failures.
llvm-svn: 55966
of RHSs to id type instead of void* if either has Objective-C object
type.
- This ensures the result can still be used in normal places an
object can be used, like a message send.
Add implicit conversions for ?: applied to qualified id types to
ensure that the RHSs are compatible.
- This prevents a codegen crash (creating invalid PHI nodes).
- Again, this relates to the fact that qualified id types have no
canonical types.
- Note that the implicit type casted to is incorrect, however this
doesn't currently cause problems because of the flexibility of the
id type.
Test cases for above.
llvm-svn: 55346
- UsualUnaryConversions takes an Expr *& and may modify its argument,
this broke when it was refactored into Sema::CheckCastTypes. This
meant that we were missing implicit casts in some places.
- Seems pretty sad that this got through our tests.
llvm-svn: 55039
- Returns addr of constant for argument + '\0'.
- I couldn't think of a better name.
- Move appropriate users of GetAddrOfConstantString to this.
Rename getStringForStringLiteral to GetStringForStringLiteral.
Add GetAddrOfConstantStringFromLiteral
- This combines GetAddrOfConstantString and
GetStringForStringLiteral. This method can be, but is not yet, more
efficient.
Change GetAddrOfConstantString to not add terminating '\0'
- <rdar://problem/6140956>
llvm-svn: 54768
- Change hardcoded assembly instructions to a use of 'opt' (still
waiting for someone to yell at me for this).
- Add some cases to cover behavior fixed by previous commit.
llvm-svn: 54413
type.
- This generates somewhat less optimal code than before but this is
not hard to rectify once stable (at the cost of slightly more
complex code).
- This currently always uses little-endian ordering of the bitfield.
- This breaks the CodeGen/bitfield.c test because it was grepping for
hard-coded assembly instructions. Will fix once a better test case
is constructed (hard to do without execution).
- This fixes SingleSource/UnitTests/2006-01-23-InitializedBitField.c
and Regression/C/PR1386.c from the test suite.
- <rdar://problem/6085090>, <rdar://problem/6094169>
llvm-svn: 54395
- Changed EmitSub to work on all cases (arithmetic, ptr - int, ptr -
ptr) so that implementation matches that of other operators.
- Modify EmitCompoundAssign to compute and perform the appropriate
casts of left, right, and result types for the assorted pointer
arithmetic cases.
- Fix EmitSub (ptr - int) case to negate the rhs *after*
extension. This is critical when the rhs is unsigned (and needs
extension).
- This fixes cfrac.
- <rdr://6115726>
llvm-svn: 54392
scimark2 on Darwin.
- Added Sema support for asm-label on variables, which I forgot before.
- Update CodeGen to use GlobalDeclMap to determine if static Decls
require emission (instead of LLVM module name lookup). Important
since the Decl name and the LLVM module name can differ.
- <rdar://problem/6116729>
llvm-svn: 54388
having multiple bindings from all the possible decls which
conceptually map to the same global).
- This eliminates CodeGen depending on the LLVM module for name
lookup.
- This also eliminates the need for ReplaceMapValuesWith (hurrah).
- This handles lookups for FunctionDecls correctly in the presence of
aliases, this was previously broken.
- WIP: Can still clean up & unify variable and function emission.
llvm-svn: 54382
- Follows emission scheme used by llvm-gcc, i.e. invent an id for
each label whose address is taken and replace each indirect goto by
a switch to each possible target.
- Currently we emit a switch for each indirect goto instead of
merging them as llvm-gcc does.
llvm-svn: 54318
- There is an miscompilation issue remaining due to a poor
interaction between the delayed emission of static functions and
the emission of constructors, but that already existed prior to
this change.
llvm-svn: 54258
instead of mapping the decl to a bitcast of the global to the correct
type.
- GetAddrOf{Function,GlobalVar} introduce the bitcast on every use now.
- This solves a problem where a dangling pointer could be introduced
by the RAUW done when replacing a forward or tentative
definition. See testcase for more details.
- Fixes <rdar://problem/6108358>
llvm-svn: 54211
- Fix multiple issues with the way case ranges were emitted, see test
cases details about the specific issues.
- The root issue was not being careful about how basic blocks were
emitted which led to them being chained together incorrectly,
resulting in improper control flow.
- Fixes <rdar://problem/6098585>
llvm-svn: 54006
used to mutate the attribute list for declspecs when the type was
converted, breaking the case where one declspec was shared by multiple
declarators.
This fixes rdar://6032532.
llvm-svn: 52769
qualifier in the lvalue, and changes lvalue loads/stores to honor
the volatile flag. Places which need some further attention are marked
with FIXMEs.
Patch by Cédric Venet.
llvm-svn: 52264
pointer cast hack currently in isIntegerConstantExpr
(in lib/AST/Expr.cpp). Also removes an odd test that even gcc doesn't accept.
The reason the pointer cast hack is relevant here is that it makes Sema
end up misinterpreting the relevant expression as a null pointer constant.
The reason for this patch is that I plan to remove the pointer cast hack
sometime soon because it causes strange issues, especially in its
current form; see my recent email to cfe-dev
"[PATCH] add constant expression evaluation to the AST and fix PR2413".
llvm-svn: 52120
reported on cfe-dev by Cédric Venet.
Note that I seriously doubt that this perticular construct is useful,
though: it's a pointer in an alternate address space pointing into
unqualified address space.
llvm-svn: 52076
this does is reconstruct the type for structs and arrays if the type
wouldn't be compatible otherwise.
The assertion about packing in the struct type reconstruction code
sucks, but I don't see any obvious way to fix it. Maybe we need a general
utility method to take a list of types and alignments and try to construct an
unpacked type if possible?
llvm-svn: 51785
nothing fundamentally wrong with it. Emitting unpacked structs where
possible is more work for almost no practical benefit. We'll probably
want to fix it at some point anyway, but it's low priority.
The issue with long double in particular is that LLVM thinks an X86 long
double is 10 bytes, while clang considers it for all purposes to be
either 12 or 16 bytes, depending on the platform, even in a packed
struct.
llvm-svn: 51673
encountered. Mixing up the decls is unintuitive, and confuses the AST
destruction code. Fixes PR2360.
Note that there is a need to look up the characteristics and
declarations of a function associated with a particular name or decl,
but the original swapping code doesn't solve it properly.
http://lists.cs.uiuc.edu/pipermail/cfe-dev/2008-May/001644.html is one
suggestion for how to fix that.
llvm-svn: 51584
they were causing bad code to be emitted. There are two fixes here: one
makes sure we emit a string that is long enough, and one makes sure we
properly handle string initialization in init lists.
llvm-svn: 51259
used for _Bool is not the same as the primitive width (which for _Bool
is 1 bit). The load and store changes add some casts to make the
types consistent. The EmitLValue changes make sure that the pointer is
of an appropriate type for loading the bitfield.
This isn't perfect, but it's an improvement, and getting everything
right depends on actually laying out structs in an ABI-compliant way.
llvm-svn: 51224
expressing the full flexibility of the LLVM shufflevector instruction.
The expected immediate usage is in *mmintrin.h, so that they don't
depend on the mess of gcc-inherited (and not completely implemented)
shuffle builtins.
llvm-svn: 51113
First, we got several CVR propagation cases wrong, which Eli pointed
out in PR2039.
Second, we didn't propagate address space qualifiers correctly, leading
to incorrect lowering of code in CodeGen/address-space.c.
Third, we didn't uniformly propagate the specifier in the array to the
pointer ("int[restrict 4]" -> "int *restrict").
This adds an ASTContext::getArrayDecayedType member that handles the
non-trivial logic for this seemingly simple operation.
llvm-svn: 49078
Eli Friedman